on oscillators

[Martin Czech]

The last thread about current mirrors, hf oscillations etc. etc. made me think...
the fewer components an oscillator core has the less problems one may get.

Triangle oscillators are nice, because they can invert direction (FM),
OTOH you need a bipolar current source, or a mirror, this is
the CA3080 app. note topology, dozens of designs come from there.

Is there any way arround the current mirror and the pnp-expo converter?

1.:
One possibility is to use two sawtooth cores in a alternating fashion,
thus hiding the discharge time. One is an inverting, the other a not inverting
integrator, so the npn current source has to be switched to one of both
integrators, but not mirrored. 

-capacitor matching (waveform symmetry), could be "trimmed" with additional
 small caps
-delay time for non inverting and inverting integrator could be different
-switching and discharge logic is more "expensive"
-sawtooth like sync need logic
-frequency wheel reversal means not to discharge, but to track, again
 extra logic
-charge injection due to switching (is there no charge injection using an ota??) 
 
This dual integrator concept is used in integrated CMOS precision clocks,
so it seems to work, the complexity of the circuit and capacitor matching
is no problem there.

2.:
The next one is even more complicated, it has only one timing cap,
and 4 switches determine which side of the cap is loaded, and which is
the reference.

-charge injection
-complicated switching
-even more complicated sync and frequency wheel reversal


3.:
The next is simpler, we know it all, it is the generic sawtooth.
What could be improoved: MOSFET source follower/MOSFET current source buffer.
Simple, very fast comparator, PMOS discharge switch.
This should insure complete and very fast discharge to the reference.

-min/max levels could vary with frequency and temperature

There are comparators with good drift specs (bipolar input stage), so
this shouldn't be too bad. It may happen that the comparator switches too
late, i.e. the lowest point gets deeper as frequency increases.
Say the saw has a 5V jump, the current drain would then happen at 50000V/s
@ 10kHz, or 50mV/us,... this is very slow even with old comparator design like
LM106 or so...

A peak-hold device could be used to determine the optimum slicing level
for the triangle and square wave converter, I've seen such a circuit.

But sawtooth oscillators can not reverse frequency wheel...
no they can. If you compare the saw wave with the resulting triangle wave
(aktive rectifier, or switching rectifier using the square output), you'll see that
a wheel reverse really means a jump to the inverse sawtooth level.
This means modifying the discharge reference level, but no extra component
in the oscillator core.

Some weeks ago there was a thread about triangle frequency wheel reversal
and it turned out to be difficult when the point of normal operation
reversal was near. The basic problem will not go away using the saw/set
feature, but maybe there is a clever way to fix it.

It seems that the good old relaxation saw oscillator can do anything
the triangle can do, including phase reversal and phase reverse sync
by simply switching the discharge reference, i.e. the simple vco
core hasn't to be changed.

Before that I was in favour of triangle osc. but after the reported problems,
and problems in finding a good pnp current source or avoiding ota/mirror problems
I'm not so sure about it any more. Especially not if the saw is needed
for timing reference (sequencers, time discrete level network).

m.c.